workflow-doctor

Where do your GitHub Actions minutes go? Reads real runs through the GitHub REST API (read-only) and reports:

• time per step (top sinks),
• job-time spent in re-run attempts,
• critical path per workflow, from the real needs in the workflow files at each run's commit,
• billed cost, using per-minute prices from config.yaml (source and check date inside).

Not implemented: cache-miss analysis (the jobs API has no cache data; it would need step logs), Markdown/HTML report, PR comment, scheduled Action.

Run

pip install -e .
export GITHUB_TOKEN=...            # any token with read access; needed to avoid the 60 req/h limit
python -c "from workflow_doctor.cli import app; app()" fetch OWNER/REPO --runs 500
python -c "from workflow_doctor.cli import app; app()" report
python -c "from workflow_doctor.cli import app; app()" critical OWNER/REPO

Example (actions/checkout, 500 runs fetched 2026-10-18; 1060 jobs):

jobs: 1060  billed cost: $16.39  (prices checked 2026-10-18)
re-run time: 7.5 min
   138.9 min  Dependabot / Run Dependabot
    62.0 min  Analyze (javascript) / Perform CodeQL Analysis

The cost is hypothetical for public repositories. The prices are from the GitHub docs page in config.yaml; that page gave no effective date.

Caveats

• Job-to-YAML matching is by job key or name:; reusable-workflow jobs and unmatched names are skipped.
• Tests use a synthetic fixture shaped like the API, not a captured real response.

Test

pip install pytest && python -m pytest -q